Diabetic Retinopathy Detection using Texture Features and Ensemble Learning

Diabetic Retinopathy, one of the dominant causes of vision loss to millions worldwide, can be prevented by early detection through regular retinal screening. People in less developed areas do not have adequate access to proper screening system because of their financial limitations. A cost-effective computer-aided screening system is presented in this paper using retinal fundus image. Ensemble learning helps to enhance the accuracy of the system by combining predictions of several learning models. In addition, these models are trained on texture features derived from gray level co-occurrence matrix (GLCM) as they are more effective to determine patterns from any images. Publicly available MESSIDOR fundus image dataset is used for experimental validation and the final results show that voting-based ensemble learning method with texture features achieves 97.2% sensitivity, 78.6% specificity and 92.0% accuracy which is higher than any individual learning model.